A framework based on evolutionary algorithm for strategy optimization in robot soccer

In any competitive and uncertain environment, designing an optimal strategy is a challenging task. The manual hand-coding of strategy is a tedious job, and its evaluation on all possible situations becomes even more complicated. This paper proposes a novel distributed framework, named FEASO, based on evolutionary algorithms, for strategy optimization in the domain of robot soccer. In the context of robot soccer, strategy denotes the critical areas where home team agents should be positioned. The focus of this study is to optimize the strategic placements of agents that are defending the goal. The presented approach comprises three modules: evolutionary algorithm execution, parallel fitness evaluation and fitness computation. It executes matches in parallel on different machines for fitness evaluation. The fitness function takes into account three parameters: the goal difference, regions occupied by defending players and ball possession by the home team players. The framework has been successfully implemented in our 3D soccer simulation team that participates in RoboCup event. Experiments are conducted using binaries of various teams taking part in the competition. A comparison of strategies between teams is conducted and analyzed. The results clearly demonstrate that the team that executes optimized strategy is able to defend more goals as compared to the team with hand-coded strategic points.